# Mouse-Geneformer Writer: Keita Ito ## Abstract This repository contains the source code of mouse-Geneformer for analysing Single-cell RNA-sequence data of mouse. mosue-Geneformer is a model pre-trained on the large mouse single-cell dataset mouse-Genecorpus-20M and designed to map the mouse gene network. The mosue-Geneformer improves the accuracy of cell type classification of mouse cells and enables in silico perturbation experiments on mouse specimens. [[bioRxiv](https://www.biorxiv.org/content/10.1101/2024.09.09.611960v1)] ## Citation If you find this repository is useful. Please cite the following references. ```bibtex @article{Ito2024.09.09.611960, author = {Ito, Keita and Hirakawa, Tsubasa and Shigenobu, Shuji and Fujiyoshi, Hironobu and Yamashita, Takayoshi}, title = {Mouse-Geneformer: A Deep Leaning Model for Mouse Single-Cell Transcriptome and Its Cross-Species Utility}, journal = {bioRxiv}, year = {2024}, URL = {https://www.biorxiv.org/content/early/2024/09/13/2024.09.09.611960} } ``` ## Enviroment Our source code is based on mplemented with PyTorch. Required PyTorch and Python version is as follows: - PyTorch : 2.0.1 - Python vision : 3.8.10 ## Execute Example of Pretraining run command is as follows: #### Pretraining ```bash # mosue-Genecorpus-20M dataset ./start_pretrain_geneformer.sh ``` Downstream task is executed in jupyter files (`cell_classification.ipynb` and `in_silico_perturbation.ipynb`) ## Trained model We have published the model files of mouse-Geneformer. `mouse-Geneformer` is base model. `mouse-Geneformer-12L-E20` is large model. - [mouse-Geneformer](https://drive.google.com/file/d/1gM3gcc3DlNGt5bAcqHbeRxtdMktGeDEg/view?usp=sharing) - [mouse-Geneformer-12L-E20](https://drive.google.com/file/d/1xKMyFA4JJeRigcJPsU2XNyxEW25Q247u/view?usp=sharing)